Guide¶

Guide component — AI consultant for optimal G6 system usage.

Cluster: Goal & Planning | Type: component | MCP Tools: 26

Overview¶

Lightweight routing and query-answering block that acts as an AI consultant for optimal G6 system usage. Records all incoming queries in in-block state and responds with structured GuideOutput answers, delegating routing logic and richer advisory operations to an inner guide_mcp layer with 26 tools covering component recommendations, usage patterns, feedback, and system navigation.

When to use:

• Answering "which G6 component should I use?" questions programmatically within a pipeline
• Routing operator queries to the right documentation, MCP tool, or component at runtime
• Embedding a thin advisory layer that can be replaced or upgraded without changing callers

Example:

from mvp.guide import GuideBlock
from mvp.guide.guide_mcp.schema import GuideInput

block = GuideBlock(name="guide")
result = block.infer(GuideInput(op="guide_ask", query="Which block should I use for web scraping?"))
# result.ok → True; result.value → GuideOutput with answer, count, success

Works well with: navigator, goal_engine, recursive_architect

Public API¶

GuideBlock(AIBlock['GuideInput', 'GuideOutput', dict])¶

Lightweight guide that answers routing queries using in-memory state.

Methods:

infer(data: Any) -> Result¶

list_patterns() -> dict[str, object]¶

Return the applied deterministic-reliability pattern catalog.

GuideClassifier¶

Fast O(log n) task classifier using decision tree over 7D epistemic vectors.

Methods:

classify(vector_array: list[float]) -> tuple[str, float]¶

Classify a 7D epistemic vector. Returns (cluster, confidence).

classify_by_keywords(query: str) -> tuple[str, float]¶

Rule-based keyword classification (zero training data).

fit(vectors: list[list[float]], labels: list[str]) -> bool¶

Fit decision tree on labeled training data. Returns True if successful.

is_fitted() -> bool¶

EpistemicVector¶

7-dimensional epistemic fingerprint.

Methods:

to_array() -> list[float]¶

from_array(arr: list[float]) -> EpistemicVector¶

neutral() -> EpistemicVector¶

EpistemicHash¶

Hash tasks and components to 7D epistemic vectors, then match.

Methods:

hash_task(query: str) -> EpistemicVector¶

Convert a natural-language task description into a 7D vector.

find_nearest(task_vector: EpistemicVector, k: int = 5, metric: str = 'l2') -> list[tuple[str, float]]¶

Return the k nearest component names with distances.

get_vector(component: str) -> EpistemicVector | None¶

Look up the pre-assigned vector for component.

get_or_default(component: str) -> EpistemicVector¶

Return the vector for component, inferring a default if unknown.

GuideMCPBlock(AIBlock[GuideInput, GuideOutput, dict])¶

Full-featured guide with 26 MCP operations.

Methods:

infer(data: Any) -> Result[GuideOutput]¶

GuideIndex¶

Searchable TF-IDF index of G6 components, tools, and clusters.

Methods:

get_corpus() -> list[dict[str, Any]]¶

Return the indexed corpus entries.

find_by_name(name: str, entry_type: str | None = None) -> list[dict[str, Any]]¶

Find corpus entries by name, optionally filtered by type.

build(components_root: Path | None = None) -> None¶

Scan components, tools, and clusters to populate the index.

search(query: str, top_k: int = 10) -> list[tuple[dict[str, Any], float]]¶

Cosine similarity search. Returns (metadata, score) descending.

boost_query(query: str, score: float) -> None¶

Record positive feedback for a query to influence future ranking.

get_boosts() -> dict[str, float]¶

Return accumulated query boosts from feedback.

stats() -> dict[str, Any]¶

Return index statistics.

GuideInput(BaseModel)¶

GuideOutput(BaseModel)¶

GuideStore¶

Constructor:

Methods:

save_template(name: str, description: str, steps: list[dict[str, Any]], effect_budget: dict[str, Any] | None = None) -> str¶

get_template(name: str) -> dict[str, Any] | None¶

list_templates(limit: int = 50) -> list[dict[str, Any]]¶

delete_template(name: str) -> bool¶

log_query(query: str, vector: list[float], results: list[dict[str, Any]]) -> str¶

save_training_data(query: str, vector: list[float], cluster: str, feedback: str) -> str¶

get_training_data(limit: int = 500) -> list[dict[str, Any]]¶

save_component_vector(component: str, vector: list[float]) -> str¶

get_component_vectors() -> dict[str, list[float]]¶

cache_explanation(key: str, explanation: str) -> str¶

get_explanation(key: str) -> str | None¶

get_recent_queries(limit: int = 20) -> list[dict[str, Any]]¶

Return recent queries from query_log, newest first.

count_all() -> dict[str, int]¶

MCP Tools¶